Re: tlug: PCMCIA Network trouble (D-Link DE-660)

[Frank Bennett <bennett@example.com>]

Hmm.  I wonder why trouble always seems to strike on weekends
when the experts are away from the office?  My first posting on
this was too long-winded for anyone to have stuck with it to the
punch line, so a quick recap:

I'm trying to back up a laptop that has to be booted via LILO off
the hard disk to a filesystem in memory, so that I can back up
the disk's data across a network.  I've decided to adapt tomsrtbt
for use as an "initrd" (for those who haven't worked with this
one, it's a ramdisk that can be loaded when Linux boots; it is
mounted as root, read-write, and persists as long as a program or
script called "linuxrc" continues to run; it can then either be
cleared from memory before mounting a physical root device like
/dev/hda1, or it can be mounted as a final root device in
memory).

The idea for the final configuration will be that the laptop
boots, performs some of the initial checks (like running e2fsck
on the local partitions) and then prompts me for the possibility
of backing up the hard disk data in one of several ways, before
proceeding to mount /dev/hda1 read-write and firing up the rest
of the system.

That's what I want to do, and the strategy seems to be working
out, up to a point.  But I'm having trouble getting the PCMCIA
card services package to bring up a working network interface
during the "linuxrc" part of the session --- and without that, I
can't get at the backup device across the network.

Tracing my steps back, it occurs to me that the problems began
when I changed to a kernel different from that used in tomsrtbt.
This was necessary because (oddly) the tomsrtbt kernel is
compiled without initrd support.

I have experimented with various combinations of the kernel
modules used by the PCMCIA subsytem, all without success.  But
through all of these trials, I have stuck with the version 2.9.12
"cardmgr" that shipped with tomsrtbt, because it (and everything
else in tomsrtbt) is linked against libc5, while my TL compiler
and later-model, known-to-work-with-my-network-card 3.0.5
"cardmgr" is linked against glibc.

I am now working on the hypothesis that ALL of the card services
package (not just the modules) has to be compiled with access to
the current kernel in order to work properly.

So ... I have broken out an old Libretto 20 that has an old
libc5-based Linux compiler in it, shifted my kernel sources and
card services sources to it, and am now recompiling the kernel,
its modules and the card services stuff on it.  When that's
finished (quite a long time from now, at 75mhz), I'll shift the
kernel into position for use in the backup entry in
/etc/lilo.conf, and move its modules and the card services
utilities into the appropriate locations in the initrd disk
image.

Then I'll be able to make a clean attempt using drivers, daemons
and a kernel that I know will work following a normal boot.  If I
still have problems after that, I'll report to the list.  For
that matter, if f things work correctly after that, I'll report
to the list.

Is anyone interested in having this stuff if I get it working
correctly?  It seems to me that a lot of laptops won't boot from
floppy; have I missed something, or is it difficult to do a
proper backup of such a machine?

Cheers,
-- 
-x80
Frank G Bennett, Jr         @@:
Faculty of Law, Nagoya Univ () email: bennett@example.com
Tel: +81[(0)52]789-2239     () WWW:   http://rumple.soas.ac.uk/~bennett/
----------------------------------------------------------------
Next Nomikai: 20 November, 19:30   Tengu TokyoEkiMae 03-3275-3691
Next Technical Meeting: 12 December, 12:30 HSBC Securities Office
----------------------------------------------------------------
more info: http://tlug.linux.or.jp Sponsors: PHT, HSBC Securities